This study describes COPD patients on regular ICS therapy who underwent a period without evidence of ICS prescription of at least 6 months. Of patients who had at least 12 months of ICS use prior to withdrawal, 11,093 (14.4%) withdrew ICS with approximately one-third receiving maintenance therapy at some point during the period following withdrawal. The remaining (65%) of patients that did not receive maintenance therapy in the period following ICS withdrawal may include patients with mild or improving symptoms that needed minimal intervention. In addition, these patients may have been prescribed a short-acting muscarinic antagonist (SAMA) and or short-acting beta-agonist inhaler (SABA) inhaler to control symptoms when needed. In addition, GOLD classification and exacerbation rates between patients receiving maintenance therapy compared to those not receiving maintenance therapy may indicate that this cohort overall had the less severe disease when contrasted to patients receiving inhaled maintenance therapy.

Overall, patients that received maintenance therapy had a longer predicted and observed withdrawal time than those without maintenance therapy and were marginally less likely to restart ICS. Routine COPD review recording coinciding with the start of maintenance therapy and a previous asthma diagnosis were key predictors of time to restarting ICS in the multivariable model. The major concern when withdrawing ICS is the possibility of triggering an exacerbation, but in this population, we found that most patients withdrawing ICS did not experience an exacerbation in the study period.

Despite the higher probability of being prescribed LAMA during withdrawal, patients that received dual therapy (LAMA/LABA or LAMA + LABA) were most likely to benefit with a longer observed ICS-free time compared with patients receiving monotherapies. Although NICE guidelines recommend dual therapy when ICS-based regimens are not needed, the finding in this study provides evidence of varied prescribing behaviour in primary care but further underscores the value of LAMA and LABA dual therapy as a more favourable maintenance treatment32.

Patients had a longer period without ICS when routine COPD review was recorded at the start of ICS withdrawal and maintenance therapy was prescribed, compared with patients without a maintenance therapy prescription. Annual patient COPD review with the GP is suggested within the primary care of Quality and Outcomes Framework33, but a GP-coded review cannot conclusively indicate that a formal COPD review with the patient took place. Alternatively, it is possible that the codes reflect the GP performing a review of the notes and prescriptions. Despite this possible rationale, COPD review codes were strongly and independently associated with ICS-free time suggesting a possible planned withdrawal and re-evaluation of a current treatment or supportive therapy. Additionally, these codes may also suggest that more reviewing of COPD medication is taking place in primary care.

Although this study was able to demonstrate the varying lengths of ICS-withdrawal period by patient characteristics and exposure to maintenance therapy, the electronic medical records did not include the reason for clinicians to initiate maintenance therapy or ICS re-initiation. As observed, just over one-half of patients receiving maintenance therapy had COPD exacerbations prior to the start of the maintenance therapy. This observation may suggest a confounding by indication bias such that clinicians maybe more likely to prescribe maintenance therapy in those they consider at higher risk of future exacerbations34. Consequently, the association between maintenance therapy and clinical outcomes during withdrawal must be interpreted with caution as the directionality of the association cannot be confirmed in this study. The effect of escalating or changing treatment class on withdrawal duration, COPD exacerbations, hospitalisations, and pneumonia episodes was not tested in this study but may provide insight as to how patient outcomes are linked with prescribing behaviours of physicians. Lastly, at baseline, the severity of COPD disease based on FEV1% was unknown in approximately one-half of all patients without maintenance treatment, which may indicate variability of how spirometry results are documented and coded. This variation in practice may result in a possible misdiagnosis of COPD or a possible misrepresentation of airflow limitation severity.

Based on the conclusions of recent meta-analyses, ICS withdrawal does not significantly increase the overall risk of COPD exacerbations; there are differences regarding FEV1 decline and quality of life metrics, which are on average below the minimal clinically important difference35,36. However, the authors note that current evidence does not evaluate the impact of ICS withdrawal after clustering COPD patients with regard to phenotype characteristics (e.g. frequent exacerbators, emphysema- hyperinflation or COPD with an asthma component), the rate of exacerbations/year (i.e.<1; ≥1 and ≤2; >2), the decline of lung function (rapid decliners vs. slower decliners) or the quality of life37,38,39. Recent studies examining primary and secondary care electronic records in the UK and Germany have also found that withdrawal of ICS is not associated with an increased risk of exacerbations compared to a cohort of COPD patients who continue ICS therapy40,41,42. A specific UK study using a large administrative healthcare database comparing a cohort of patients withdrawing from a triple-therapy ICS-based regimen also suggested no increase in moderate and severe exacerbations due to a higher number of pneumonia consultations recorded in primary care at baseline42. The current study adds further information as to the way ICS therapy is withdrawn in a primary care setting. Although a minority of patients received a COPD review when prescriptions for ICS were ceased, these patients were least likely to restart ICS, which suggests that withdrawing ICS may be a part of an ICS-step down plan, an approach that is recommended by The Primary Care Respiratory Society (PCRS)43. In this cohort, patients that experienced exacerbations were also more likely to be on maintenance therapy, but it is not possible to draw conclusions from this observation as there may be confounding factors that were not considered, unclassified disease severity groups at baseline, and a likely reverse causality between the need for maintenance therapy and clinical outcomes. For example, in the Copenhagen General Population Study44, adherence to maintenance medication for COPD was low, although this increased with the progressive severity of the disease as defined by the GOLD stage.

This study has several strengths to understand the characteristics of patients and associated periods of ICS withdrawal. We utilised data from a large, well-validated primary care database with linkage to secondary care data. The identification of acute exacerbations of COPD from electronic health records is well-validated, and since it is based on data from routine care, the prescription data are likely to be entered correctly as the electronic health record system is required to generate a patient prescription30,45. This study provides a good capture of general prescribing behaviour in the COPD patient population as these patients are predominately managed in primary care.

Nevertheless, we cannot know if the prescription was dispensed, nor whether the patient took the medication, and adherence to maintenance medication in COPD is known to be poor44. In this study, we were interested in the effects of ICS withdrawal, so we can be relatively confident that the absence of a prescription record indicates that medication has not been supplied. It is likely that these patients were prescribed short-acting bronchodilation therapy or supportive care therapies for the relief of both respiratory and non-respiratory symptoms respectively, suggestive that patients had regular or routine visits to the GP to manage symptoms during the withdrawal period, however; we cannot confirm that all cases of withdrawal are genuinely deliberate attempts to withdraw ICS at the direction of a GP. For some patients, experiencing unusually longer gaps between withdrawal date and re-uptake of ICS therapy could be attributed to failure to refill prescriptions (i.e. patient-driven withdrawal) or stockpiling of previously issued prescriptions of ICS monotherapy or ICS combined therapy with LAMA or LABA. Although previous studies have described how to minimise the risk of steroid withdrawal22,23,24,25, this study seeks to describe a steroid withdrawal population in practice. The comparison of patients in this study to those that did not withdraw ICS for a minimum of 6 months may provide insight on COPD exacerbations and outcomes, but we cannot guarantee a sufficient patient population of continuous long-term ICS use in addition to the historical 12-month persistent ICS use prior to index date. Patients are more likely to start and stop ICS perhaps due to changing clinical signs or higher exacerbations events compared to those on long-acting therapies only. For this reason, comparison ICS withdrawn patients to ICS persistent users may lead to differential misclassification of outcomes thus weakening the association between clinical outcomes and ICS withdrawal.

This study has also shown that many patients in the withdrawing population had an historic or concurrent asthma diagnosis and the decision to withdraw may be confounded by the coded diagnosis of asthma rather than COPD, with asthma diagnosis possibly over-recorded in individuals with COPD46. This may explain why concomitant asthma increased the risk of restarting ICS, as these patients were being managed more as asthma patients than COPD, although severity criteria specifically for asthma (e.g. Asthma Control Test score) were not captured in this study. Historical asthma diagnosis was also more prevalent in patients who did not receive maintenance therapy, which may suggest that these patients have stepped off ICS due to improvement of asthmatic symptoms.

Results from the IMPACT39 and SUNSET47 clinical trials suggest that the risk of exacerbations following ICS withdrawal, although small, primarily occurs in the first 28 days37 and the step down from LAMA/LABA in combination with ICS-based regimen to LAMA/LABA dual maintenance therapy show no significant difference in COPD exacerbations, especially in patients with blood eosinophils <0.3 × 109 cells/L45. Although the IMPACT trial has highlighted that sudden step down from triple therapy to dual maintenance therapy can introduce exacerbations, it may be due to the fact these patients were frequent exacerbators and characterised with asthmatic features who may have otherwise benefited from continued ICS treatment. Overall, these findings suggest that withdrawal must be carefully investigated on a case-by-case basis based on multiple patient-based factors. Our definition of withdrawal included a minimum 6-month ICS-free period and therefore could not specifically identify the immediate increased risks of ICS- withdrawal within this study, and therefore direct comparison with IMPACT and SUNSET is not possible. Our data provided insight into characteristics and longer-term outcomes of patients that were not restarted on ICS due to short-term reactions to withdrawal.

Whether withdrawal of ICS occurs in a planned or unplanned way, the majority of COPD patients withdrawing ICS did not experience exacerbations and COPD-related hospitalisations. When ICS-withdrawal occurs in a planned way with a COPD review taking place and maintenance therapy prescribed, the ICS-free period is longer compared to patients without a prescribed maintenance therapy or a recorded COPD review. This may result in the reduction of side effects and enable more cost-effective prescribing. This study provides evidence that ICS withdrawal results in minimal exacerbations of COPD, thus supporting the guidance on recent ERS ICS withdrawal21. Withdrawal success can be optimised by carefully planning COPD reviews whilst optimising the use of available maintenance therapies.